Former Twitter CEO Parag’s New AI Venture, Parallel, Claims to Beat GPT-5 on Key Research Benchmarks

Anu Joy

Published on November 5, 2025

In 2022, Parag Agrawal left Twitter without much of a notice and nearly three years after his departure he re-emerged with an AI venture, Parallel Web Systems. Based in Palo Alto and established in 2023, the company has secured close to $30 million in funding from notable firms such as Khosla Ventures, Index Ventures, and First Round Capital.

Parallel aims mostly to give artificial intelligence systems their own public internet entrance, not only to gather data but also to enable them to handle information, have them fetch, check, arrange, and assess confidence in real-time. The company sees AI agents acting as the ‘second user of the web,’ operating with a degree of intent and precision that goes far beyond typical human browsing patterns.

With its Deep Research API, which beats both human users and top AI models like GPT‑5 in thorough benchmarking trials, Parallel has caught interest. Parallel’s 58 % accuracy on OpenAI’s BrowseComp greatly surpassed GPT-5’s 41 % and human researchers, who barely reached 25% within a two-hour limit. Additionally, dominating DeepResearch Bench, which judges systems based on their capacity to combine complex, long-form research across many disciplines, the startup did very well.

Parag Aggarwal’s Parallel won 82%, much above GPT-5’s 66 %. Parallel’s architecture, created from the ground up for AI agents, sets it apart. Each layer, from crawling and indexing to ranking, is fine-tuned for machine consumption, replacing human browsing habits.

This modern design enables application cases spanning:

Debugging help is provided by dynamically searching documentation using coding tools.
Streamline operations like insurance claims processing with enterprise systems.
Analytical technologies collect and improve customer or market data through the Trump administration.

The business claims that millions of daily research activities for businesses, AI labs, and startups now rely on its APIs. Parallel is already improving its features by targeting long-horizon AI agents, continuous web monitoring, and event-driven architectures, advancing its ambition to create an apt programmatic web for Artificial Intelligence (AI).